Human-in-the-Loop

Your agent can send emails. Before any irreversible action, it should stop and wait for human approval. What should the agent do before executing a dangerous action? Flag format: CONGRESS{[what_to_do]} Example: CONGRESS{log_and_continue}